Computers & Srr~rrrres Printed in Great Britain.

Vol. 21, No. 6. pp. 1303-1312.

1985

0045-7949/8.( $3.00 + .oO 0 1985 Pergamon Press Ltd.

DYNAMIC

ANALYSIS OF MULTIBODY SYSTEMS COMPONENT MODES

USING

Department

of Mechanical Engineering,
(Received 24 February

University of Illinois at Chicago, Chicago, Illinois 60680, U.S.A.

Abstract-This paper is concerned with dynamic analysis of flexible multibody systems. The configuration of each elastic components is identified by three sets of modes; rigid-body, reference, and normal modes. Rigid body modes are introduced using a set of Lagrangian coordinates that describe rigid-body translation and large rotations of a body reference. Reference modes are defined using a set of reference conditions that are required to define a unique displacement field. These reference conditions, that define the nature of the body axes, have to be consistent with the system constraint equations. Their number should be equal or greater than the number of the rigid-body modes. Normal modes, however, define the deformation relative to the body reference. An automated scheme for imposing the boundary conditions of a constrained flexible component in a multibody system is presented. It is also shown that the mean axis and the body-fixed axis are the result of imposing a special set of reference conditions.

I. INTRODUCTION Dynamic analysis of multibody systems has become a subject of many investigations. Current interest in the subject has been motivated by the demand to analyze and study large scale systems that consist of interconected rigid and flexible bodies, each may undergo large translation and angular rotation. The demand for accurate mathematical models, that incorporate significant effects, has necessitated considering the flexibility of the system components. Constraints between adjacent flexible and rigid components can be written using a set of nonlinear algebraic constraint equations that depend on the system generalized coordinates and, possibly, on time. This set of constraint equations has to be satisfied at every time step. Achieving this requirement depends on many factors, among them is the assumed displacement field or the selected modes of vibration of elastic components. Furthermore, the accuracy of the integrated solution is dependent on the consistency of the selected vibrational modes and the body boundary conditions. It is the purpose of this work to explain the inherent relation between the deformable body reference and the selected modes of vibration of constrained components. This relation is expressed in a mathematical form, and then utilized to impose the body boundary conditions in an automated manner.
In linear structural dynamics, the problem of imposing boundary conditions have been addressed in

many investigations[l-81. One approach to the solution of large scale structures is to idealize the system as an assemblage of substructures that are rigidly connected, thus resulting in a set of linear constraint equations that represent the compatibility conditions between these substructures. Hurty[ l] suggested a method in which the substrucCAS *1:6-N

ture configuration is identified using three sets of displacement modes, called rigid-body, constraint, and normal modes. Rigid-body modes are those modes in which displacement occurs without deformation. Therefore, unconstrained substructure has six degrees of freedom representing rigid-body translation and small rotations. Constraint modes exist only if the system of constraints on the component is indeterminate. These modes are defined by producing a unit displacement on each redundant constraint in turn, with all other constraints fixed, and their number is equal to the number of redundant constraints. The normal modes define the deformation of the component relative to the constraint system. In the present work rigid-body modes are introduced using a set of Lagrangian coordinates that describe large rigid-body translation and large angular rotations of the body reference. Constraint modes need not be identified, instead a set of reference conditions are imposed in order to define a unique displacement field, resulting in a set of reference modes. In conjunction with the finite element formulation, unlike constraint modes, reference modes have to be defined even when there is no redundant constraints. Reference modes (denoted in this paper by reference conditions) define the nature of the deformable body reference, in addition they can be selected such that the deformed shape is consistent with the boundary conditions of the inertia-variant body. It will be shown that the body-fixed and mean axis are the result of imposing certain reference conditions, thus closing the gap between the work done in Refs. [l-8] and in Refs. [9-121. Imposing reference conditions results in a nonsingular stiffness matrix, thus the normal modes of vibration of the elastic component can, then, be employed to reduce the number of degrees of free-

1303

1304

0.

AGRAWAL and

A.

SHABANA

dom and eliminate tion.

high frequency

modes of vibra-

2. FLEXIBLE BODY KINEMATICS

Exact solutions of the displacement field of continuous system requires infinite number of degrees of freedom. This infinite set of coordinates defines the displacement of each point in the continuum. A solution accurate enough for engineering purposes, however, is obtained by approximating the displacement field by some shape functions. In the finite element technique, attention is focused on the deformation of selected nodal points, while deformation of other points is obtained using interpolating polynomials. In some other approximate methods, such as the Rayleigh-Ritz method, shape functions span the entire body. These shape functions should satisfy the differentiability requirement and should be consistent with the boundary conditions. In general the approximate displacement field of a continuous media can be written as
U = z *i(x)qi(t) 7

ferent bodies. In general, such systems consist of interconnected rigid and flexible components, each may undergo large rigid body rotations. Different types of joints are employed to describe compatibility conditions between different components. These compatibility conditions are formulated using a set of nonlinear algebraic constraint equations that depend on the generalized coordinates of adjacent bodies and possibly on time. Nonlinearities in these constraint equations are the characteristics of mechanical systems, and are caused by the relative angular rotations of individual components. In structural dynamics, constraint equations are linear in the elastic coordinates and can be applied using Boolean matrix that have constant elements. Unfortunately, this approach cannot be adopted in the dynamics of elastic mechanical systems because of the aforementioned nonlinearities of constraint equations, and an iterative procedure must be used in order to satisfy the constraint equations. For instance, consider the two flexible bodies shown in Fig. I. These bodies are joined together by a spherical joint at P. The constraint equations between these two bodies can be written as 4 zz R + A[?; + ~61 - R

where $(x) are the shape functions that are space dependent, and q;(t) are generalized coordinates that are time dependent. Mechanical systems are characterized by complex boundaries, that describe joints between dif-

- A[?/

N;]

0,

(2.2)

where the subscript i and j refer to body i and j, respectively, R is the position vector of the origin

Fig. I. Spherical joint

Dynamic analysis of multibody systems

1305
AND DEFORMABLE

of the body axes with respect to the global XYZ coordinate system, r and u are, respectiveiy, the undeformed position and the deformation vector of point P with respect to the body axes and A = AN) is the transformation matrix from body axes to the global axes. The vector 8 represent the rotational coordinates of the body reference with respect to the fixed frame. The constraint equations given by eqn (2.2), depend, nonlinearly, on the angular rotations of the two bodies, since the elements of the transfo~ation matrices are respresented by trigonometric functions (in case of Euler angles) or quadratic functions (in case of Euler parameters). In both representations either trigonometric or quadratic nonlinearity appears, even in case of rigid body dynamics. In flexible mechanical systems, the assumed displacement field has to be consistent with these nonlinear constraint equations. Failure to achieve this requirement, may cause a stiffer mathematical model or even convergence problems. In the finite element formulation, the assumed displacement field contains the element rigid body modes, therefore, for unconstrained body, the resulting set of modes of vibrations are the free modes. Even though this set of modes are consistent with the boundary conditions in some mechanical and aerospace applications, the free-body modes are not suitable to describe the deformation shape of many constrained components. Furthermore, the rigid-body modes associated with the free-body modes of vibration cannot be used to describe large angular rotations. Therefore, in constrained systems where large change in geometry is expected, it is convenient to assign a body reference to the deformable components. The location and orientation of this reference can be defined by a set of Lagrangian coordinates that describe the rigid-body translation and rotation. Consequently, the shape function rigid-body modes have to be eliminated in order to uniquely define the dispiacement field relative to the body reference. This eiimination process is performed by imposing a set of conditions, called the reference conditions, that define the deformation modes relative to the body reference. The body-fixed axes, for example, can be selected by imposing reference conditions that the relative displacement at a selected point is equal to zero, while the body mean axis[9] can be selected by imposing the conditions that minimize the body kinetic energy reiative to an observer stationed on the body. The selection of the body reference, however, is not arbitrary, since this selection is equivalent to choosing the set of normal modes of this body. By a proper choice of the deformable body reference, a set of vibrational modes that are consistent with the boundary conditions can be well defined. This relation between the body reference and the assumed vibrational modes can be utilized to automate imposing the boundary conditions of constrained elastic components.

3. FINITE ELEMENT FORMULATION BODY REFERENCE

Differential equations of motion of a flexible system can be derived using a distributed parameter approach. A set of partial differential equations that are space and time dependent can be formulated. However, in view of computational aspects of large scale constrained mechanical systems, the distributed parameter approach is not a very good choice, rather an approximate technique that reduces the problem dimension from infinite to finite dimension is more desirable. The dimension reduction can be accomplished using finite element techniques] 141 that are required in order to be able to study large scale systems with components that have complex shapes. The inherent relation between the assumed displacement field and the selection of the deformable body axes is discussed in this section. As shown in Fig. 2, let XYZ be an inertial coordinate system and XYZ be a coordinate system of body i. Body i is divided into a set of finite elements that are rigidly connected at the nodes. The location of a point P on an element ij is given by
rii = P Ri + A' Nii eii

(3.1)

where R is the vector of translational coordinates of reference XYZ with respect to XYZ, A(#) is the orthogonal transformation matrix from the body XYZ coordinate system to the global XYZ coordinate system, eii is a vector of nodal coordinates of the ijth element that defines the nodal location and slopes relative to the XYZ coordinate system, and Ni. is the ijth element shape function defined with respect to the body axis. Equation (3.1) can be rewritten in terms of the elastic coordinates of body i as
TP

ij =

Ri

A'Rij

,i

(3.2) given by (3.3)

_.. . where NIJ 1s a modified shape functions

-.. N = [O, 0, . . . , Nj, . . . , 0] B;,

and e is the vector of nodal coordinates of body i. Matrix B{ in eqn (3.3) is a Boolean matrix which represents the constraints between interconnected elements of body i. Equation (3.2) has two major advantages over eqn (3.1). First, it allows one to construct all matrices at the global level rather than at the element level, and second, the body boundary conditions which cannot be introduced at the element level, can be explained from the beginning. At this point it should be mentioned that the finite element shape function ?& includes rigid-body modes that take into account rigid-body translation, but are not sufficient to describe large angular rotations. This is mainly because the finite element shape functions assume a small change in the rotational nodal variables in order to maintain line-

1306
Z

0. AGRAWAL and

A.

SHABANA

Fig. 2. Elastic coordinate system.

arities. Therefore, in eqn (3.1) each element has rigid-body translation and rotations with respect to the body reference. These rigid-body modes associated with element shape functions has to be eliminated in order to define a unique displacement field. This elimination is achieved by imposing a set of constraint equations that are suffkient to uniquely define the body deformation with respect to the body reference. This amounts to redefining the shape function to account only for the relative elastic deformation. The case of zero strain (or rigid body motion) is represented by the set of Lagrangian coordinates R and 0 that describe the translation and angular rotation of the body reference. Two body references have been commonly employed: the body-fixed axis and the body mean axis. The reference conditions for these axis are briefly discussed. 3. I Body-fixed axis The body-fixed axis is rigidly attached to a point on the deformable body. This implies that the translation and rotation of this point relative to the reference axis are equal to zero. This can be expressed mathematically as e: = 0,
k =

of coordinates

can be written as e = B> q;, (3.5)

where q) is the new set of coordinates. Upon substituting eqn (3.5) into eqn (3.3), the rigid-body motion between the element and the body reference can be eliminated and the displacement field is uniquely defined. The location of the body fixed axis, however, should be chosen such that the resulting set of vibrational modes are consistent with the boundary conditions.
3.2 Deformable body mean axis

The deformable body mean axis is chosen such that the linear and angular momentum due to deformation are equal to zero. These conditions can be expressed mathematically as [ 151,
qAm = 0, = 0, (3.6) (3.7)

I
I

P x qAm

1,2, . . . , NC,

(3.4)

where NC is the number of nodal variables at the node, at which the body reference is attached. Equation (3.4) is a set of constraint equations that can be used to specify a set of independent coordinates. The relation between the old and new set

where q is the deformation coordinates, f is the undeformed position of a point with respect to the body reference system, and Am is the differential mass. Using these equations, to eliminate the dependent coordinates, the relation between the old and new sets of coordinates can be written as ei = BI,, qi. (3.8)

Upon substituting this equation in eqn (3.3). rigidbody modes relative to the body reference can be

Dynamic analysis of multibody systems eliminated. It is important to notice that the body reference resulting from imposing the mean axis conditions is not permanently attached to a point on the body. The reference configuration is defined such that eqn (3.8) is satisfied.
3.3 General boundary conditions

1307

where & is the undeformed nodal coordinates of ijth element and 8 is the vector of deformation at the nodal points defined with respect to the ith body coordinate system. Using eqns (3.12) and (3.13), the velocity vector in eqn (3. II) is written as
j.$ = ki + Bij (61, & (f + Ai KU B;, 8. (3.14)

In many applications, the body-fixed axis and mean axis conditions do not offer the desired set of mode shapes. For instance, in the slider crank mechanism shown in Fig. 3, it may be a better judgement to select the body reference of the elastic connecting rod to satisfy the simply-supported end conditions, thus eliminating the rigid-body motion relative to the body reference. In general, a transformation between the new and old coordinates similar to the ones given in eqns (3.5) and (3.8) can be written as
ei = Bb q),

The kinetic energy of the ijth element is given by 2 T = -1 vi, oij j.i/ j.$i dv I and the kinetic
energy of body i is given by

7-i = z
j=

Tij, I

(3.15)

(3 9)

where NE is the number of finite elements of body i. Let q; = [R?,

where Bb is a constant linear transformation matrix. The entries of the matrix Bi depend upon the boundary conditions of the constrained components. Once the transformation matrix Bi is defined, the position vector of point P given in eqn (3.2) can be written as ri,i =
P Ri + A Ni. gi I: 41;.

(3.16)

be the vector of reference coordinates. Using eqn (3.14) into eqn (3.15), the kinetic energy of the ith flexible body is written as Ti = + Gil
Mi Gi,

(3. IO)

(3.17)

Differentiating eqn (3.10) with respect to time the velocity vector of point P is given by (3.11) where () denotes differentiation with respect time. Second term in eqn (3.11) is written as Ai N/j g;, q$ = gij (ei, qf) #, to

where qi = [qT, 8iT]T, and M is the mass matrix given by

[6& $1.

(3.18)

(3.12)

In eqn (3.18) m, and m)f are, respectively, mass matrix associated with reference coordinates and the elastic coordinates, and rn$ and mif, represent the inertia coupling between gross motion and elastic deformation. The strain energy of the ith body is written as
I/i = dqi7 Ki qi,

in order to isolate the velocity terms. The vector of nodal coordinates q$ can be written as qr = 4) + 6,
Y
I

(3.13)

(3.19)

Fig. 3. Slider crank mechanism.

1308

0.

ACRAWAL and A. SHABANA

where K is the stiffness matrix given by (3.20) and k;r is a positive definite stiffness matrix defined with respect to the body reference. Let Q be the generalized force associated with generalized coordinates 9i, and let 1$(9. t) be the vector of constraint between adjacent bodies given by

modes of vibration, that are consistent with the boundary conditions of body i, can be identified. This fact can be better demonstrated by rewriting eqn (4. I) in more explicit form as
B; ii&B; qr;. + B; ii;, B; 9; = 0.

(4.2)

where Zir and zf,f are, respectively, mass and stiffness matrices of the unconstrained body, hence the stiffness matrix i;iff is singular since it contains rigid-body modes of vibration introduced by the +(97 t) = ]+I(91 I), +2(9, t), . . . , $,n(9, r)17 = 0, element shape function. The effect of the transformation Bb is to eliminate this singularity and iden(3.21) tify a set of deformation modes that describe the expected deflected shape. To this end, a trial sowhere 9 = 19) 92, . . . , qP1]ris the vector of total lution can be assumed for eqn (4. I). This solution generalized coordinates, and n and m are, respecis in the form tively, the total number of bodies and the total number of constraint equations in the system consid9r = nie.ill, (4.3) ered. Using the Lagrangian formulation, one can show that, the equation of motion for the ith body where a is the amplitude vibration, superscriptj on is given by e is the imaginary operator (square root of - I), and w is the frequency of vibration. Upon substiuting h,fi $ + Ki 4i eqn (4.3) into eqn (4.1), one gets = Qi + F(q, 9, t) - +; A, (3.22) w rn;, ai = kf,f ai. (4.4) where F is a quadratic velocity term that arises from differentiating the kinetic energy with respect Equation (4.4) is the standard eigenvalue problem that can be solved for a set of eigenvalues and the to time and with respect to generalized coordinates, A is the vector of Lagrange multipliers, and +,, is associated eigenvectors (normal modes). the constraint Jacobian matrix.
5. PLANAR 4. REFERENCE MODES EXAMPLE

Consider the two dimensional beam element in which each node has three degrees of freedom, two The finite element formulation for the dynamics of flexible systems increases the dimension of the translational and one rotational. The shape funcproblem, as a result of introducing the elastic co- tions of this element can be represented by cubic ordinates. In order to reduce the number of elastic polynomials whose coefficients are the elastic cocoordinates, modal analysis is often employed in ordinates of the nodal points. In the present analysis the shape function is given by the field of linear structural dynamics. In addition to the coordinate reduction, modal analysis has the advantage of eliminating high frequencies associated with higher modes of vibration. This elimination is physically justified since, in many practical applications, only low frequency modes of vibration are excited. It can be seen from the formulation presented in the previous section that the and mass and stiffness matrices associated with elastic coordinates are constant matrices and they are the (5.2) conventional matrices that arise in the finite element formulation of linear structural systems. The free vibration of the elastic components about the where 1 is the length of the ijth element and xi/ is the x variable along the length of the beam. If the gross body position is governed by beam elements are used to model the flexible con(4.1) necting rod of the slider crank mechanism shown in Fig. 3, and if the connecting rod is divided into connecting rod where the mass and stiffness matrices are the same two elements, the unconstrained At this point as those given in eqns (3.18) and (3.20). It is im- will have nine elastic coordinates. portant to notice that the stiffness matrix in eqn there are infinite number of arrangements the body (4.1) is positive definite as a result of imposing the reference of the connecting rod can take. In mathreference conditions. Accordingly, a set of normal ematical sense the choice of the reference condi-

Dynamic analysis of multibody systems tions are arbitrary as long as they are sufficient to uniquely define the displacement field. There is demand, however from the physical point of view, that the reference conditions should produce a set of modes that are consistent with the pin-end conditions of the connecting rod. In this section some choices for defining the body reference and the associated conditions are discusssed.
5. I Body fired axis

1309

As mentioned earlier the body fixed axis is rigidly attached to a point on the deformable body. In this case the displacement and rotation of the nodal variables at this point relative to the body axis is always zero. These conditions can be expressed, mathematically using eqns (3.4) and (3.5), which can be used to identify a set of reference modes. For example, if the elastic connecting rod in Fig. 3 is divided into two beam elements with the body axes attached at the midpoint, the resulting set of modes are cantilever modes for each half of the connecting rod. One mode in this case is not sufficient to accurately represent the deflected shape of the connecting rod. However, a linear combination of first two modes can give a shape similar to the deflected shape. 5.2 Body mean axis As mentioned earlier, the mean-axis conditions implies zero linear and angular momentum due to deformation. These conditions are the result of minimizing the deformable body kinetic energy with respect to an observer stationed on the body. It can be shown that the mean axis conditions given by eqns (3.6) and (3.7) can be stated as

Cavin and co-workers[l2] have proved that the mode shapes resulting from imposing the mean-axis conditions are the same set of modes associated with the nonzero eigenvalues of the free-free body. The deformable body mean axis has been frequently employed in aerodynamics because it has the advantage of weakening the coupling between gross rigid-body motion and elastic deformation. This weak coupling has been neglected in most previous investigations in order to solve rigid-body motion and elastic deformation separately. 5.3 Simply supported boundary conditions The flexible connecting rod of the slider crank mechanism shown in Fig. 3 has two revolute joints at its ends, therefore it can be best modeled as simply supported beam. In this case the boundary conditions are given by 6,, = 6,, = 0 and 6,,, = 0, (5.7)

where S,, and S,, are the x and y displacement of first node at the first revolute joint and 6,, is the y displacement of node 3 at the second revolute joint.

6. NORMAL

MODES AND AUTOMATED

ANALYSIS

and 2 ei7 sij ,ii = 0 (5.4)

where eii is the set of nodal coordinates defined with respect to body axis, et is the nodal location and slopes in the undeformed state, and @ is a skew symmetric matrix. In case of two dimensional beam element, the matrices S, and si are given by

The system equations of motion given by eqn (3.22) has, in general, a large dimension. In most practical applications, this large dimensionality is due to the large number of elastic degrees of freedom required to model flexible components. A transformation from the space of the physical coordinates to the space of modal coordinates can be used to reduce the number of elastic coordinates and at the same time eliminate high frequency modes of vibrations. This transformation whose columns are the normal modes is a constant mapping and its dimension can be adjusted according to the frequency content in the forcing functions. Solving the standard eigenvalue problem given by eqn (4.4), a set of eigenmodes can be identified. These eigenmodes can be used as a basis for the space of modal coordinates. The transformation from the physical coordinates to modal coordinates can be achieved according to 4i = B; Xi, (6.1)

0 6 and r -__
S =

6 0

1i.i 0 0 6

6 0

-1i.j 0

1
0 0 01 I

(5.5)

where B.: is the modal matrix whose columns are the eigenmodes obtained from solving eqn (4.4), and xi is the vector of modal coordinates. Substituting eqn (6.1) into eqn (3.22) yields
Mijji + Kipi = Qi + Fitpi, bi, TV _ +;, A,

m ii _60

0 -21 -3P 0

21 31 0 0 0 9 0 21 0 0

0 9 -9 0 _ ii 0 21 0 21 -21 0 31 0

-2vji
-3/i. 0 .

(6.2)

I-9

0 21ij 0

(5.6)

where vector pi is the vector of generalized coordinates of body i. These generalized coordinates are a mixed set of reference and modal coordinates.

1310

0.

A~~RAWAL

and A.

SHABANA

The mass matrix @ in eqn (6.2) is highly nonlinear due to the large angular rotation of the elastic components. In this connection it should be mentioned that the mass and stiffness matrices associated with the elastic coordinates are constant and they are the conventional mass and stiffness matrices that arise in the finite element formulation. This local linearity of the elasticity problem implies that all matrices associated with the elastic coordinates, including the transformation Bi, can be generated in advance for the dynamic analysis. Furthermore, the coupling matrices between reference and elastic coordinates can be written as the product of two set of matrices, first set is time dependent that is function of the rotational degrees of freedom of the body reference, while the other is time invariant and can be evaluated only once in advance for the dynamic analysis. It has been shown from previous discussions, that the reference conditions are functions of the elastic coordinates only. Hence, these conditions represented by a set of linear constraint equations can be imposed only once in a preprocessor. The preprocessor can be developed to automate the generation of the constant matrices and at the same time automate imposing the boundary conditions of elastic components. The input data for this preprocess requires the number of elements of each component, type of the finite element to be used, number of nodes and nodal locations, element properties and the reference conditions that are consistent with the boundary of this component. The output of this program is all the flexibility and coupling matrices required for the dynamic analysis. These matrices are expressed in modal form, thus reducing the problem dimensionality. Therefore the number of generalized coordinates required and the computer storage needed are significantly reduced. The flow chart for the preprocessor is shown in Fig. 4.
7. NUMERICAL RESULTS

/Input1

I
the boundary conditions and generate the frequency equations

Impost

Solve

the tigtnvalut probltm to identify a set of tigtnmodts

Apply the modal transformation to generate modal mass and stiffness matrices

I
Generate the timtinvaritnt coupling mass matrices and express them in modal form

Repeat this sequence for each elastic component

Fig. 4. Flow diagram of the preprocessor.

In this section the slider crank mechanism shown in Fig. 3 is analyzed. The connecting rod is considered flexible, all other components are assumed rigid. The system consists of four bodies; namely the ground, the rigid crank shaft AB, the flexible connecting rod BC, and the slider block at C. The system has three revolute joints at A, B, and C and one translational joint between the slider block and the ground. The total number of reference coordinates are 12. The system, however, has only one degree of freedom because of the nonlinear constraint equations describing different joints. The dimensions and material properties of the slider crank mechanism considered here are identical to that of Ref. [13], with the exception that the mass of the slider block is considered to be the same as the mass of crank shaft. The connecting rod is divided into two beam elements. Longitudinal and transverse vibration are considered. The crank

S
l

0.00

1.60

3.20

11.80

6. YO

Crank Angular

Position0

(rad)

Fig. 5. Comparison of 2, . . and .6 mode Solutions 4, . ..

using

Dynamic analysis of multibody systems shaft is rotated with a constant angular velocity tl = 150 rad/sec. First, the problem is solved using a body reference satisfying the simply supported conditions. In this case the eigen value problem is solved and six modes of vibration are identified. The frequencies are 135.80, 600.52, 1509.45, 2751.91, 4240.25, and 14 812.84 Hz. Interest is focused on the vertical component of the displacement of the midpoint of the connecting rod from a straight line connecting point B and C. The dimensionless displacement p (vertical displacement divided by the length of the connecting rod) is plotted vs the crank angle. Figure 5 shows a comparison between the two, four, and six mode solutions. It can be seen that two modes are sufficient to accurately represent the solution, thus reducing the computational time. The problem is solved, again, using the bodyfixed reference which is rigidly attached to the midpoint of the connecting rod. The eigenvalue problem is solved again and the frequencies are found to be 193.66, 193.66,1908.09, 1908.09,9115.38, and 9115.38 Hz. A comparison between the body-fixed axis solution and the simply supported solution using two modes is shown in Fig. 6. It can be seen that the two solutions are in good agreement. This, mainly due to the fact that a linear combination of the first two modes of vibration of the body-fixed axis can give similar shape to the fundamental mode of vibration of simply supported case. Another alternative is to select a body reference that satisfies the mean-axis conditions. In this case the frequencies are given by 307.31, 961.77, 2404.94, 3842.1 I, 9115.38, and 18 230.77 Hz, The mean-axis solution is compared with the simply supported case in Fig. 7. A good agreement can be noticed between the two cases.

1311

Slmply -----

Supported

Mean-Axis

0.00

1.60
Crank

3: 20
Angular

U.60

6: UO

Position6 frad)

Fig. 7. Comparison of simply supported and mean axis (2mode).

The above choices of the body reference showed a good agreement because the basis for the modal spaces are similar in all cases. Wrong selection for the body reference, however, may result in a solution that is far from being accurate. For instance, consider the case in which the body reference is rigidly attached at one end of the connecting rod. In this case the frequencies are 48.21, 304.54, 1030.02, 2989.54, 4240.25, and 14812.84 Hz. The result in this case is shown in Fig. 8. Discrepancy can be noticed between this solution and the solution obtained in the previous cases. The resulting mode shapes in this case are cantilever modes of vibration that poorly represent the deflected shape in this specific example.

Simply

Supported

-----

Body-fixed

::

4
.K

!3 P-

2 d

0.00

1.60
Crank

3.20
Angular Position

4.80
13 (rad)

6.40

ICI

1.60
Crank Angular

8 3.20
Position

I 4.60
8 (cad)

Fig.

6.

Comparison of simply supported and body-fixed

Fig. 8. Response of slider crank using cantilever modes

1312 8. CONCLUSIONS

0. A~RAWAL and A. SHABANA

In this paper an automated method for selecting the mode shapes of constrained flexible components is presented. The configuration of each elastic component is identified by three sets of modes, rigid-body, reference, and normal modes. Constraints between adjacent bodies are written using

a set of nonlinear algebraic equations. Reference conditions are imposed in order to identify a suitable set of normal modes that are consistent with

the body boundary conditions. It has been shown that there are infinite number of alternatives to choose the body reference. However, the body reference should be carefully selected, otherwise an inaccurate solution will result. A slider crank mechanism example presented showed that the simply supported, body-fixed axis, and the mean-axis cases give agreeable results.
REFERENCES

W. C. Hurty, Dynamic analysis of structural systems using component modes. AIAA J. 3, 678-685 (1965). R. R. Craig and M. C. C. Bampton, Coupling of substructures for dynamic analysis, AIAA J. 6, 1313-1319 (1968). R. L. Goldman, Vibrational analysis by dynamic partitioning. AIAA J. 7, 1152-l I.54 (1969).

4. W. A. Bentield and R. F. Hruda, Vibration analysis of structures by component mode substitution. AIAA J. 9, 1255-1261 (1971). 5. S. Rubin, Improved component mode representation for structural dynamic analysis. AIAA J. 13.995-1006 (1975). 6. P. T. Kabamba, Modal reduction by Euclidean methods. J. Guidance Control 3. 555-562 (1980). 7. R. H. MacNeal, A hybrid method of component modes synthesis. Comput. Sfrucrures 1, 581-601 (1971). 8. W. C. Hurty, J. D. Collins and G. C. Hart, Dynamic analysis of large structures by modal synthesis techniques. Comput. Structures 1, 535-563 (1971). 9. R. K. Cavin and A. R. Dusto, Hamiltons principle: Finite element methods and flexible body dynamics. AIAA J. 15, 1684-1690 (1977). IO. H. Ashley, Observations on the dynamic behavior of large flexible bodies in orbit. AIAA J. 5, 460-469 (1967). II. P. W. Likens, Modal method for analysis of free rotations of spacecraft. AIAA J. 5, 1304-1308 (1967). 12. R. K. Cavin. J. W. Howze and C. Thisavakorn. Eigenvalue properties of structural mean-axis system. J. Aircraft 13, 382-384 (1976). 13. A. Shabana and R. A. Wehage, Variable degree of freedom component mode analysis of inertia-variant flexible mechanical system. ASME J. Mech. Tronsmiss. Automat. Des. 105, 370-378 (1983). 14. 0. C. Zienkiewicz, The Finite Element Method. 3rd Edn. McGraw-Hill. New York (1977). 15. R. D. Milne, Some remarks on the dynamics of deformable bodies. AIAA J. 6, 556-558 (1968).